Autofocus Method for SAR Based on Dynamically Randomized Block Sampling in Wavenumber Domain

Synthetic Aperture Radar (SAR) systems on small unmanned aerial vehicles (UAVs) are challenged by complex, spatially-variant motion errors, which conventional autofocus methods often fail to adequately correct. Specifically, approaches using fixed sub-block partitioning struggle to adapt to the non-uniform spatial distribution of phase errors, while static sample selection mechanisms lack the flexibility to optimize estimation throughout the iterative process. To overcome these limitations, this paper proposes a novel autofocus algorithm that integrates two key innovations. The first, adaptive slant-range wavenumber sub-block partitioning, dynamically adjusts the division granularity based on the local severity of the phase error, ensuring an optimal trade-off between estimation accuracy and robustness. The second, a sample selection strategy guided by a Dynamic Probability Density Function (DPDF), adaptively modifies sample selection probability during iterations, balancing broad exploration in early stages with focused exploitation of high-quality samples in later stages. Integrated within a fast factorized back-projection (FFBP) framework, the method leverages the coherence between azimuth phase error (APE) and non-systematic range cell migration (NsRCM) for joint estimation and correction. Validation using both simulated and measured UAV SAR data demonstrates that the proposed method significantly enhances focusing quality.